A history lesson, before we get to the Stew rant
Secondary school, a WW1 module, a teacher called Davies. He was working his way toward explaining why Britain abandoned decades of "splendid isolation" to enter into alliances it had spent a generation avoiding. I hadn't turned the page. I blurted out, entirely too loudly for a classroom, "balance of power." Davies stopped. He knew it was the right answer. I knew it was the right answer. The textbook, unhelpfully, hadn't got there yet. "Did you read ahead?" he asked, or words to that effect.
I hadn't. What I had done, repeatedly, obsessively, was read The Godfather. It's the only book I've ever finished and immediately flipped back to page one to start again. The whole thing opens on the idea that no great fortune exists without a crime sitting somewhere underneath it, an idea the book borrows, more or less, from Balzac. I could once tell you, to the page, where in the old grey Brando-cover edition Tom Hagen and Michael have their conversation about nothing personal, just business (p.148 I think). It's been about fifteen years since I last picked up a copy and none of that has gone anywhere.
So I explained it to Davies the only way I actually understood it at fifteen: Vito Corleone doesn't move against Sollozzo because Sollozzo has done something to him. He moves because if Sollozzo's heroin business is allowed to grow unchecked, there comes a point where the Corleone family stops being able to choose the moment of confrontation at all. Best case, they get absorbed into someone else's order. Worst case, they get destroyed at a time of somebody else's choosing. Splendid isolation ends for the same reason. You don't wait until the other side is strong enough that you no longer get a vote.
And the Balzac line matters here too, not just the Corleone one. Behind every great fortune, a crime, or at least a decision someone would rather not have examined too closely. Keep that in your back pocket, because it's basically the subtext of everything that follows.
I bring this up now because it's the only honest lens I've got for the last fortnight , and readers marking the 4th this weekend might enjoy watching a country currently very fond of its own sovereignty use exactly this logic on a company of its own.
9 June: launched anyway
Anthropic released Claude Fable 5 to the public, less than two months after warning, when it gave Mythos a limited private preview in April, that the underlying model's capabilities had drawn "significant concern from technology, finance, and government leaders." Fable and Mythos are, per Anthropic, essentially the same model, differentiated by safeguards and access tiers; both can operate unattended on user commands for longer stretches than any previous Claude release. The launch landed as Anthropic's private valuation approached $1 trillion.
If you're feeling the Balzac itch already, hold that thought. Every great fortune, etc.
12 June: the switch gets pulled
Two days after general availability, at 5:21pm Eastern on 12 June, Anthropic received an export control directive ordering it to suspend Fable 5 and Mythos 5 access for any foreign national, anywhere, including its own foreign-born staff. With no way to verify nationality in real time, Anthropic pulled both models for everyone rather than risk non-compliance; every other Claude model was unaffected. The government gave no detail on the underlying national security rationale beyond pointing to a guardrail bypass technique. Anthropic pushed back hard, saying its own review found the capability in question already achievable on other publicly available models, including GPT-5.5, and standard practice for cybersecurity defenders.
Read next to the Corleone logic above, this is Washington doing the Vito Corleone move rather than the Sollozzo one, acting before a capability tips into something it can no longer contain, not after. Whether an hour's notice on a Friday evening was a proportionate way to do that is a fair question, and not one this fortnight's coverage answered convincingly either way.
The timing bit commercially too: this landed weeks after Anthropic disclosed a $47 billion revenue run rate, a $965 billion valuation, and a confidential IPO prospectus already filed with the SEC.
26 June–1 July: the climbdown
The order lifted on 30 June. Fable 5 access was restored across Claude Platform, Claude.ai, Claude Code and Claude Cowork on 1 July, ending an eighteen-day suspension. Mythos 5 came back more cautiously, limited for now to a set of US organisations approved by the government on 26 June, with international Project Glasswing partners still pending.
More interesting for governance purposes: Anthropic used the moment to propose a joint severity framework for jailbreaks, developed with Amazon, Microsoft and Google under the Glasswing banner, scoring techniques on capability granted, how broadly it applies, ease of weaponisation, and how discoverable the technique already is. The stated goal is a consistent, objective standard the industry currently lacks, so both developers and governments have a shared basis for triage. Anthropic is inviting other providers to adopt it.
2 July: the safeguards get a proper airing
Anthropic followed up with detail on how Fable 5's cybersecurity classifiers actually work day to day, sorting requests into four bands: prohibited (malware, ransomware), high-risk dual use (penetration testing, privilege escalation, blocked pending better requester verification), low-risk dual use (OSINT, mostly allowed), and benign (debugging, patch management, not meant to be caught at all). The aim on vulnerability discovery specifically is blocking Fable 5 from finding flaws other widely available models can't, while leaving standard defensive security work alone.
The Cyber Jailbreak Severity scale floated alongside this is the same four-factor model teased on 1 July, now formalised into five bands running Informational to Critical, developed with Glasswing partners and open for feedback from academia, industry, civil society and government before Anthropic treats it as settled.
Meanwhile, on identity: 1984 gets a corporate makeover
Sitting slightly outside the Fable 5 timeline, but landing in the same fortnight, Anthropic's revised Privacy Policy takes effect Wednesday coming. The relevant addition is a new "Verification Data" category: if a user is asked to verify their age or identity, and chooses to do so, Anthropic can collect an image of their government-issued ID, their image in photo or video form, and facial geometry templates, which the policy itself flags as data that "may be considered 'biometric data' in some jurisdictions" depending on where the user lives. Worth being precise here: the policy frames this as something Anthropic "may ask" for "in certain circumstances," tied to age or identity checks, not a blanket requirement rolled out to every user on every login. The legal basis Anthropic cites for collecting it is consent, alongside legitimate interests, legal obligation, and, in some cases, vital interests. The policy doesn't apply to Enterprise customers, whose data is governed separately under their own commercial agreements.
On retention, the policy is genuinely thin. It commits to holding personal data only "for as long as reasonably necessary," pointing to a privacy centre for further detail, without publishing a fixed schedule for how long a face or an ID scan specifically gets kept.
Orwell's line about Big Brother watching you has been doing overtime for decades as shorthand for state surveillance, but by 2026 the more accurate version probably swaps out the state for the tech stack sat in your living room. It isn't the government reading your messages that keeps most people up at night, it's Anthropic, Meta, Google, Amazon, all quietly deciding what counts as reasonable data collection in a privacy policy nobody reads past paragraph two.
There's an old joke that's aged uncomfortably well on this point: a man tells his wife he's been speaking softly around the house because he's worried Mark Zuckerberg is listening in. She laughs. He laughs. Alexa laughs. Siri laughs. Nobody in that room finds it reassuring for very long, and the joke only works because everyone already half-believes it's true.
Also this fortnight: Anthropic starts making drugs
Smaller item, but worth a line: Anthropic announced it's moving into drug development itself via a new application called Claude Science. STAT's report quotes life sciences lead Eric Kauderer-Abrams framing it as wanting hands-on experience solving "real scientific problems" with Anthropic's own tools, rather than a straightforward commercial push, and it remains unclear whether any candidate is actually intended for market.
